Advanced Integration-Inspired Process-in-Memory: A Comprehensive Review of Design, Challenges, and Future Prospects

Abstract

This work provides a comprehensive overview of advanced integration-inspired memory process design, focusing on the integration methods of 2D, 2.5D, and 3D. The integration of compute and memory blocks has demonstrated encouraging outcomes in enhancing memory access latency and energy efficiency for data-centric operations. The challenges and future prospects for PIM design include addressing trade-offs in the integration process, thermal management in highly integrated 3D designs, developing design automation tools for 3D designs, and identifying practical application scenarios for PIM to maximize its benefits. The use of through-silicon vias (TSVs) technology enables the efficient design of 3D structures with improved bandwidth density, while hybrid bonding technology shows advantages in achieving higher interconnect density and data transmission bandwidth. Monolithic integration represents a distinctive solution for high-density module integration, whereby custom circuit layers are stacked vertically in a single chip. This approach represents an efficient alternative to traditional 2D integration technologies. These advances in integration technology provide a roadmap for the future of PIM design and its application in real-world scenarios. In conclusion, the continuous development and integration of different technologies in PIM design has the potential to significantly enhance performance, energy efficiency, and integration density for data-centric tasks.

Keywords

• Process-in-memory
• heterogeneous integration
• near-memory computing